The Agent Skills Directory

[COMMAND_EXECUTION]: The skill requires and executes system binaries xelatex and bibtex to compile dynamically generated LaTeX source code into PDF artifacts. This execution is managed by the paper-artifact-runtime skill and includes multiple paper-latex-sanitizer steps designed to filter malicious LaTeX commands before compilation.
[EXTERNAL_DOWNLOADS]: The search_papers step utilizes the multi-search-engine skill to fetch academic metadata from well-known services like Crossref, Brave, and Tavily. The gathered data is used to construct the paper's bibliography through the refbib and source_pack steps.
[PROMPT_INJECTION]: The skill processes untrusted user input (topic and paper requirements) to drive the generation of the manuscript. To mitigate indirect prompt injection risks, it consistently applies xml_escape and truncate filters to the user message before it is interpolated into LLM prompts across various steps (e.g., paper_collect, paper_contract).
[DATA_EXPOSURE]: The skill creates and writes files to the local filesystem within a run-specific directory (paper/{{ inputs.meta_run_id }}/). The final PDF is delivered using the publish_artifact tool, which is restricted to the specific output path of the generated paper.

meta-paper-write